Former OpenAI CTO Mira Murati has returned to the scene with her new company Thinking Machines. On Monday, they introduced a concept called interaction models, which allows users to collaborate with AI in real time by continuously processing audio, video, and text. The key difference is that these systems respond like a human, without waiting for you to finish speaking or writing.
How the new continuous interaction approach works 🤖
Interaction models eliminate the bottleneck of current systems, which process inputs discretely and lose live context. Thinking Machines uses continuous attention architectures that integrate simultaneous multimodal streams. This allows the AI to react to changes in tone, gestures, or interruptions, adjusting its response on the fly. The technology is based on modified transformers designed to maintain a persistent contextual state without resets.
Now the AI also interrupts you when you speak 🗣️
Finally, an AI that doesn't wait for you to finish your sentence to give you a generic response. Now you can have heated discussions with your assistant, who will cut you off mid-sentence to tell you you're wrong. Sure, at least you won't have to repeat the context every time you breathe. The dream of arguing with a machine as if it were your brother-in-law is now a reality.